Every fast write moves work somewhere else
Summary
A technical exploration of where write work actually happens in modern storage systems. The article discusses the trade-offs between returning success after memory, local SSD flushes via fdatasync, or remote/durable storage, and how write-ahead logs, batching, and replication (including Raft) affect durability, latency, and coordination. It emphasizes that Big-O growth describes workload scaling but does not by itself capture real-world latency or reliability across different storage paths.